Scorpio MCTS+NN: What are the strongest nets?

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Rebel, chrisw

BrendanJNorman
Posts: 2526
Joined: Mon Feb 08, 2016 12:43 am
Full name: Brendan J Norman

Scorpio MCTS+NN: What are the strongest nets?

Post by BrendanJNorman »

Maybe this is best for Daniel...

Big fan of your engine lately.

Don't care how strong it is honestly, but I like playing AGAINST human-like opponents.

Scorpio has been a fun test recently.

What nets would you recommend I use?

Anyone else know?

And where to download them?
Daniel Shawul
Posts: 4185
Joined: Tue Mar 14, 2006 11:34 am
Location: Ethiopia

Re: Scorpio MCTS+NN: What are the strongest nets?

Post by Daniel Shawul »

I have never really measured its strength honestly because all I do is check if it works by running it through a set of epd positions.
I have nets trained from about 25000 games (actually set of epd positions) called nets-epd and from about 0.7 million CCRL games (nets-ccrl)
that can be downloded here https://github.com/dshawul/Scorpio/releases . There are nets starting from very small 2x32 upto 24x128

I have fixed a very serious bug with the CPU version in parallel processing and added multi-GPU support for game play yesterday. The lattter is not
very efficient yet.

Daniel
BrendanJNorman
Posts: 2526
Joined: Mon Feb 08, 2016 12:43 am
Full name: Brendan J Norman

Re: Scorpio MCTS+NN: What are the strongest nets?

Post by BrendanJNorman »

Daniel Shawul wrote: Wed Sep 19, 2018 10:24 am I have never really measured its strength honestly because all I do is check if it works by running it through a set of epd positions.
I have nets trained from about 25000 games (actually set of epd positions) called nets-epd and from about 0.7 million CCRL games (nets-ccrl)
that can be downloded here https://github.com/dshawul/Scorpio/releases . There are nets starting from very small 2x32 upto 24x128

I have fixed a very serious bug with the CPU version in parallel processing and added multi-GPU support for game play yesterday. The lattter is not
very efficient yet.

Daniel
Thanks Daniel,

I tried the 24x128 net and played against it as black, getting an advantage within about about 15 moves in a Tarrasch Defense.

Its relatively weak on my MacBook Pro CPU (I'd estimate about 2100-2200 real Elo), but plays interesting chess and will be a useful training partner.

Will follow your project with interest.

Thanks for all you do.
Daniel Shawul
Posts: 4185
Joined: Tue Mar 14, 2006 11:34 am
Location: Ethiopia

Re: Scorpio MCTS+NN: What are the strongest nets?

Post by Daniel Shawul »

BrendanJNorman wrote: Wed Sep 19, 2018 12:06 pm
Daniel Shawul wrote: Wed Sep 19, 2018 10:24 am I have never really measured its strength honestly because all I do is check if it works by running it through a set of epd positions.
I have nets trained from about 25000 games (actually set of epd positions) called nets-epd and from about 0.7 million CCRL games (nets-ccrl)
that can be downloded here https://github.com/dshawul/Scorpio/releases . There are nets starting from very small 2x32 upto 24x128

I have fixed a very serious bug with the CPU version in parallel processing and added multi-GPU support for game play yesterday. The lattter is not
very efficient yet.

Daniel
Thanks Daniel,

I tried the 24x128 net and played against it as black, getting an advantage within about about 15 moves in a Tarrasch Defense.

Its relatively weak on my MacBook Pro CPU (I'd estimate about 2100-2200 real Elo), but plays interesting chess and will be a useful training partner.

Will follow your project with interest.

Thanks for all you do.
Thanks! Note that on the CPU, the bigger the net the worse it plays.
So your best network for the CPU is probably the 2x32 nets.
BrendanJNorman
Posts: 2526
Joined: Mon Feb 08, 2016 12:43 am
Full name: Brendan J Norman

Re: Scorpio MCTS+NN: What are the strongest nets?

Post by BrendanJNorman »

Daniel Shawul wrote: Wed Sep 19, 2018 5:51 pm
BrendanJNorman wrote: Wed Sep 19, 2018 12:06 pm
Daniel Shawul wrote: Wed Sep 19, 2018 10:24 am I have never really measured its strength honestly because all I do is check if it works by running it through a set of epd positions.
I have nets trained from about 25000 games (actually set of epd positions) called nets-epd and from about 0.7 million CCRL games (nets-ccrl)
that can be downloded here https://github.com/dshawul/Scorpio/releases . There are nets starting from very small 2x32 upto 24x128

I have fixed a very serious bug with the CPU version in parallel processing and added multi-GPU support for game play yesterday. The lattter is not
very efficient yet.

Daniel
Thanks Daniel,

I tried the 24x128 net and played against it as black, getting an advantage within about about 15 moves in a Tarrasch Defense.

Its relatively weak on my MacBook Pro CPU (I'd estimate about 2100-2200 real Elo), but plays interesting chess and will be a useful training partner.

Will follow your project with interest.

Thanks for all you do.
Thanks! Note that on the CPU, the bigger the net the worse it plays.
So your best network for the CPU is probably the 2x32 nets.
Thanks Dan, I'll try the smaller nets.

Do you think it would be fun to make a net from say, all the games of aggressive world champions (Alekhine, Tal, Spassky, Kasparov) and/or a separate one based on positional world champions (Capablanca, Smyslov, Botvinnik, Petrosian, Karpov, Carlsen)?

Not so much for strength, but to see the resulting play.

This type of thing is really interesting to me - and I'm sure some others. ;)

I don't think with Scorpio you have any bold intentions to be the next Lc0 or A0, and you already have a great base engine...this "style" modelling might be fun.
Daniel Shawul
Posts: 4185
Joined: Tue Mar 14, 2006 11:34 am
Location: Ethiopia

Re: Scorpio MCTS+NN: What are the strongest nets?

Post by Daniel Shawul »

It seems that I have made a mistake in counting the number of blocks , so what I thought was a 2x32 net was actually a 3x32 net...same for other nets.
I retrained an actual 2x32 net in 10 minutes over 25000 games and recompiled then tensorflow with optimizations on.
Then this version has the same nps number as the the scorpio-mcts version without neural networks !!!
Also I am using min-max backup with mcts instead of the averaging as in the standard mcts. This has made it tactically very good
and easily destroys tscp as the first 6 games below show. Note that I am only talking about the CPU. I don't think
you will be able to beat this version unless you are >2400.

To answer your question, yes training new networks I am sure will be very fun. Also since my input planes are attack tables,
and I have separate nodes for piece counts, you can capture piece values and attack features very fast. Try it and let me know.

[pgn] [Event "Computer Chess Game"] [Site "dani-s-xp"] [Date "2018.09.20"] [Round "1"] [White "scorpio"] [Black "tscp"] [Result "1-0"] [TimeControl "40/120"] [Annotator "19. +1.90 12... -0.05"] 1. e4 e5 2. Nf3 Nc6 3. Bb5 a6 4. Ba4 Nf6 5. O-O Be7 6. Re1 b5 7. Bb3 O-O 8. c3 d6 9. h3 Na5 10. Bc2 c5 11. d4 Qc7 12. Nbd2 cxd4 {-0.05/6 4} 13. cxd4 Bd7 {+0.00/6 4} 14. Nf1 Rfc8 {-0.01/6 4} 15. Ne3 Nc6 {+0.05/6 4} 16. Nd5 Nxd5 {+0.18/6 3} 17. exd5 Nb4 {-0.06/6 3} 18. Bb1 f6 {+0.13/6 3} 19. dxe5 {+1.90/71 10} fxe5 {+0.02/6 3} 20. Bg5 {+0.64/71 5} Bxg5 {-0.04/6 3} 21. Nxg5 {+0.43/71 1.2} Nc2 {-0.09/6 2.9} 22. Ne6 {+0.72/71 5} Bxe6 {-0.06/6 2.8} 23. Bxc2 {+0.67/71 5} Bd7 {-0.19/6 2.7} 24. Rc1 {+0.80/71 5} Qa7 {-0.26/6 2.6} 25. Qh5 {+2.41/71 5} h6 {-0.49/6 2.6} 26. Qg6 {+1.35/71 5} Qc5 {-0.59/5 2.5} 27. Qh7+ {+6.27/71 5} Kf8 {-0.43/5 2.4} 28. Re3 {+6.87/71 5} Qxd5 {-0.12/5 2.3} 29. Rf3+ {+6.11/71 5} Qxf3 {-1.40/5 2.2} 30. gxf3 {+7.21/71 1.3} Bxh3 {-1.69/6 2.2} 31. Qg6 {+7.23/71 5} Rd8 {-1.60/6 2.1} 32. Bb3 {+10.59/71 6} d5 {-2.63/6 2.0} 33. Rc7 {+13.11/71 6} Rd7 {-3.13/6 1.9} 34. Rxd7 {+14.78/71 1.4} Bxd7 {-7.08/7 1.9} 35. Bxd5 {+16.21/71 6} Ke7 {-8.27/6 1.8} 36. Qxg7+ {+16.58/71 6} Kd6 {-9.29/6 1.8} 37. Bxa8 {+16.73/71 1.6} h5 {-9.76/7 1.7} 38. Qf8+ {+18.28/71 8} Ke6 {-10.65/6 1.6} 39. Be4 {+20.52/71 8} a5 {-11.43/6 1.6} 40. Bf5+ {+20.65/71 2.3} Kd5 {-12.25/6 1.5} 41. Bxd7 {+20.55/71 3} b4 {-13.10/7 5} 42. Qa8+ {+22.28/71 3} Kd6 {-13.93/6 5} 43. Qc6+ {+22.74/71 3} Ke7 {-14.73/6 5} 44. Qe6+ {+22.72/71 3} Kf8 {-14.77/6 5} 45. Qf6+ {+23.03/71 3} Kg8 {-99.92/6 1.7} 46. Qe7 {+99.75/71 3} h4 {-99.94/5 0.1} 47. Be6+ {+99.84/71 3} Kh8 {-99.96/3 0.1} 48. Qf8+ {+99.96/71 0.8} Kh7 {-99.98/2 0.1} 49. Bf5# {+99.98/71 0.8} {Xboard adjudication: Checkmate} 1-0 [Event "Computer Chess Game"] [Site "dani-s-xp"] [Date "2018.09.20"] [Round "2"] [White "tscp"] [Black "scorpio"] [Result "0-1"] [TimeControl "40/120"] [Annotator "5. +0.20 1... +0.00"] 1. Nf3 Nf6 {+0.00/72 0.1} 2. d4 e6 {+0.00/72 0.1} 3. c4 b6 {+0.00/72 0.1} 4. Nc3 Bb4 {+0.00/72 0.1} 5. Bd2 {+0.20/6 4} O-O {+0.00/72 0.1} 6. e3 {+0.25/7 4} Bb7 {+0.00/72 0.1} 7. Bd3 {+0.19/6 4} c5 {+0.00/72 0.1} 8. dxc5 {+0.26/6 4} Bxc5 {+0.13/71 7} 9. O-O {+0.23/6 3} d6 {+0.04/71 3} 10. e4 {+0.31/6 3} Qd7 {+0.33/71 3} 11. Bg5 {+0.62/5 3} Nc6 {-0.09/71 3} 12. Bxf6 {+0.89/5 3} gxf6 {-0.08/71 0.8} 13. Qd2 {+0.84/5 3} Qd8 {-0.03/71 3} 14. Qf4 {+0.94/5 2.9} a5 {+0.03/71 3} 15. Rad1 {+0.94/5 2.8} Qe7 {-0.22/71 3} 16. Rd2 {+1.02/5 2.7} Rad8 {-0.36/71 3} 17. Qg4+ {+1.04/5 2.7} Kh8 {+0.00/72 0.1} 18. Rfd1 {+1.02/5 2.6} Ne5 {+0.93/71 4} 19. Qh3 {+0.93/6 2.5} Ba8 {+0.31/71 4} 20. Nxe5 {+1.04/5 2.4} dxe5 {+0.74/71 4} 21. Qg4 {+0.89/5 2.3} Rg8 {+0.64/71 4} 22. Qh4 {+0.78/5 2.2} Rdf8 {+0.71/71 4} 23. Bb1 {+0.94/5 2.2} Bc6 {+0.45/71 4} 24. Qh3 {+0.79/5 2.1} Re8 {+1.21/71 4} 25. Bd3 {+0.80/5 2.0} Rb8 {+1.01/71 4} 26. Bc2 {+0.84/4 2.0} f5 {+0.52/71 4} 27. Re2 {+0.76/4 1.9} f4 {+1.01/71 4} 28. Red2 {+0.74/5 1.8} Rbc8 {+1.08/71 4} 29. Ba4 {+0.76/4 1.8} Bxa4 {+1.04/71 4} 30. Nxa4 {+0.56/6 1.7} Bd4 {+1.23/71 4} 31. b3 {+0.51/5 1.7} Qf6 {+1.28/71 4} 32. Qf3 {+0.58/5 1.6} Rb8 {+1.61/71 4} 33. Nc3 {+0.68/5 1.5} Bc5 {+0.58/71 4} 34. Rd7 {+0.80/5 1.5} Rbf8 {+0.20/71 4} 35. Rb7 {+0.97/4 1.4} Qh4 {+0.47/71 4} 36. Na4 {+1.73/5 1.4} Bd4 {+0.53/71 4} 37. Nxb6 {+2.20/5 1.3} Rd8 {+0.23/71 4} 38. Rc1 {+2.12/5 1.3} f5 {-0.71/71 4} 39. Rc2 {+2.15/5 1.3} Bxb6 {+4.35/71 4} 40. exf5 {-0.84/6 1.2} Qh3 {+10.01/71 4} 41. Kf1 {-2.14/7 5} Qxf5 {+4.98/71 2.9} 42. Qe2 {-1.27/6 5} Rxg2 {+8.71/71 2.9} 43. Kxg2 {-2.35/6 5} f3+ {+10.34/71 0.7} 44. Qxf3 {-4.20/7 5} Rg8+ {+10.51/71 0.7} 45. Qg3 {-4.40/7 4} Rxg3+ {+10.80/71 0.8} 46. hxg3 {-4.40/6 4} Qe4+ {+11.18/71 0.8} 47. Kg1 {-4.70/7 4} Qxb7 {+10.49/71 0.8} 48. Rd2 {-5.55/7 4} Qf3 {+12.15/71 3} 49. c5 {-7.15/7 4} Bxc5 {+13.19/71 0.8} 50. g4 {-7.78/7 4} Bb6 {+14.72/71 3} 51. g5 {-8.15/7 4} Qg3+ {+14.93/71 3} 52. Kf1 {-7.53/6 4} Qh3+ {+14.71/71 0.8} 53. Ke1 {-7.75/6 3} Qh1+ {+15.30/71 3} 54. Ke2 {-7.95/6 3} Qe4+ {+14.37/71 3} 55. Kf1 {-7.42/6 3} Bd4 {+13.77/71 3} 56. a3 {-8.20/6 3} Qh1+ {+15.25/71 3} 57. Ke2 {-8.27/6 3} Qh5+ {+14.83/71 3} 58. f3 {-8.18/6 2.9} Qxg5 {+15.08/71 0.9} 59. Rd3 {-8.20/7 2.8} Qg2+ {+15.37/71 3} 60. Kd1 {-8.87/6 2.7} Qg1+ {+15.51/71 3} 61. Kd2 {-8.97/6 2.6} Qf2+ {+15.35/71 3} 62. Kd1 {-9.02/6 2.5} Qb2 {+15.40/71 4} 63. b4 {-8.74/7 2.5} Qb1+ {+17.13/71 4} 64. Ke2 {-8.62/6 2.4} Qa2+ {+16.05/71 4} 65. Kd1 {-8.64/6 2.3} Qc4 {+15.47/71 4} 66. Kd2 {-8.73/7 2.2} axb4 {+16.07/71 4} 67. axb4 {-8.93/7 2.1} Kg7 {+15.63/71 4} 68. b5 {-9.13/7 2.1} Qxb5 {+16.71/71 4} 69. Kc2 {-9.20/6 2.0} Qb2+ {+17.31/71 4} 70. Kd1 {-10.28/5 1.9} Kf6 {+17.24/71 4} 71. Rd2 {-10.95/7 1.9} Qb3+ {+18.50/71 4} 72. Ke2 {-10.95/5 1.8} Qc4+ {+17.81/71 4} 73. Kd1 {-10.95/5 1.8} Qf1+ {+17.39/71 4} 74. Kc2 {-10.95/5 1.7} Ke7 {+17.06/71 4} 75. Rd3 {-10.90/6 1.6} Qe2+ {+17.79/71 4} 76. Rd2 {-11.00/6 1.6} Qxf3 {+18.23/71 4} 77. Rd3 {-11.09/6 1.5} Qe2+ {+18.55/71 4} 78. Rd2 {-11.09/5 1.5} Qc4+ {+19.79/71 4} 79. Kd1 {-12.38/5 1.4} Be3 {+21.22/71 4} 80. Rh2 {-12.80/6 1.4} Qd3+ {+23.70/71 1.1} 81. Ke1 {-12.70/6 5} Bf4 {+21.14/71 2.9} 82. Rg2 {-13.14/6 5} Qe3+ {+22.33/71 0.7} 83. Kf1 {-12.80/6 5} Bg3 {+20.52/71 3} 84. Ra2 {-12.72/7 5} Qf3+ {+20.23/71 0.8} 85. Kg1 {-12.72/6 5} Kf6 {+18.17/71 0.8} 86. Rb2 {-12.90/7 4} Bf2+ {+27.65/71 3} 87. Rxf2 {-5.28/10 4} Qxf2+ {+51.74/71 3} 88. Kxf2 {-4.96/12 4} h5 {+53.19/71 3} 89. Kg3 {-4.93/11 4} e4 {+52.92/71 3} 90. Kf4 {-3.66/12 4} h4 {+52.66/71 3} 91. Kxe4 {-3.66/11 4} h3 {+53.73/71 3} 92. Kf3 {-2.06/11 4} h2 {+54.04/71 3} 93. Kg2 {-2.06/12 4} h1=Q+ {+57.36/71 0.8} 94. Kxh1 {-2.06/11 3} Ke5 {+54.11/71 3} 95. Kg2 {-2.30/13 3} Kf4 {+54.51/71 3} 96. Kf2 {-2.16/13 3} e5 {+54.91/71 3} 97. Ke2 {-2.16/12 3} Ke4 {+0.00/72 0.1} 98. Kd2 {-8.70/12 3} Kf3 {+54.18/71 3} 99. Ke1 {-8.80/12 2.9} e4 {+55.06/71 3} 100. Kd1 {-8.90/12 2.8} e3 {+55.12/71 3} 101. Ke1 {-9.10/11 2.7} e2 {+0.00/72 0.1} 102. Kd2 {-9.20/11 2.6} Kf2 {+0.00/72 0.1} 103. Kd3 {-9.20/9 2.5} e1=Q {+57.86/71 1.0} 104. Kd4 {-9.20/7 2.4} Qe2 {+56.41/71 4} 105. Kd5 {-9.20/7 2.4} Ke3 {+56.56/71 4} 106. Ke5 {-9.30/7 2.3} Qg2 {+56.69/71 4} 107. Kd6 {-9.30/7 2.2} Kd4 {+56.71/71 4} 108. Ke6 {-9.40/7 2.1} Qd5+ {+99.92/71 4} 109. Kf6 {-9.40/6 2.1} Qe4 {+99.92/71 4} 110. Kf7 {-9.40/7 2.0} Ke5 {+99.92/71 4} 111. Ke8 {-9.40/6 1.9} Qh7 {+99.94/71 4} 112. Kd8 {-99.96/5 0.1} Kd6 {+99.96/71 4} 113. Ke8 {-99.98/3 0.1} Qg8# {+99.98/71 4} {Xboard adjudication: Checkmate} 0-1 [Event "Computer Chess Game"] [Site "dani-s-xp"] [Date "2018.09.20"] [Round "3"] [White "scorpio"] [Black "tscp"] [Result "1-0"] [TimeControl "40/120"] [Annotator "1. +0.00 6... -0.59"] 1. e4 {+0.00/72} c5 2. Nf3 {+0.00/72 0.1} e6 3. d4 {+0.00/72 0.1} cxd4 4. Nxd4 {+0.00/72 0.1} Nc6 5. Nc3 {+0.00/72 0.1} Qc7 6. Be3 {+0.00/72 0.1} Bb4 {-0.59/6 4} 7. Nb5 {+1.57/71 7} Bxc3+ {-0.63/6 4} 8. bxc3 {+3.12/71 3} Qb8 {-0.52/6 4} 9. Nd6+ {+3.15/71 3} Kf8 {-0.75/5 4} 10. Rb1 {+3.28/71 3} Nge7 {-0.55/6 3} 11. Bf4 {+3.22/71 3} e5 {-0.65/6 3} 12. Bc1 {+2.70/71 3} Kg8 {-0.26/6 3} 13. Bc4 {+4.06/71 3} Nd8 {-0.76/6 3} 14. Bg5 {+3.86/71 3} Kf8 {-0.69/6 3} 15. f4 {+4.42/71 3} exf4 {-0.90/6 2.9} 16. Rf1 {+4.29/71 3} Ne6 {-0.61/6 2.8} 17. Bxf4 {+4.04/71 3} Nxf4 {-0.52/6 2.7} 18. Rxf4 {+5.45/71 0.8} f6 {-1.15/6 2.7} 19. Qd4 {+7.76/71 3} Nf5 {-3.69/6 2.6} 20. Rxf5 {+10.51/71 0.9} Qxd6 {-10.71/6 2.5} 21. Qxd6+ {+19.25/71 0.9} Ke8 {-10.71/5 2.4} 22. e5 {+19.34/71 4} Kd8 {-10.94/6 2.3} 23. exf6 {+19.89/71 4} Re8+ {-14.95/6 2.2} 24. Kd2 {+19.97/71 4} gxf6 {-16.60/6 2.2} 25. Qxf6+ {+25.46/71 4} Re7 {-17.32/5 2.1} 26. Re1 {+27.55/71 4} b6 {-17.53/6 2.0} 27. Qd6 {+99.69/71 4} Rxe1 {-99.96/5 0.2} 28. Kxe1 {+99.91/71 4} Bb7 {-99.98/3 0.1} 29. Rf8# {+99.98/71 0.9} {Xboard adjudication: Checkmate} 1-0 [Event "Computer Chess Game"] [Site "dani-s-xp"] [Date "2018.09.20"] [Round "4"] [White "tscp"] [Black "scorpio"] [Result "0-1"] [TimeControl "40/120"] [Annotator "3. -0.78 1... +0.00"] 1. c4 e5 {+0.00/72 0.1} 2. g3 Nf6 {+0.00/72 0.1} 3. Nc3 {-0.78/6 4} Bb4 {+0.00/72 0.1} 4. a3 {-0.69/6 4} Bd6 {+0.26/71 6} 5. Nf3 {-0.46/6 4} b6 {-0.21/71 3} 6. d4 {-0.11/6 4} exd4 {+0.00/71 3} 7. Nb5 {-0.01/6 3} Bb7 {-0.15/71 3} 8. Nxd6+ {+0.23/6 3} cxd6 {-0.22/71 0.8} 9. Qxd4 {+0.15/6 3} Qe7 {-0.14/71 3} 10. Qc3 {+0.23/6 3} Nc6 {-0.08/71 3} 11. Bg5 {+0.41/5 3} Ne5 {+0.22/71 3} 12. Bg2 {+0.41/6 2.9} Rc8 {+0.52/71 3} 13. b3 {+0.37/6 2.8} h6 {+0.26/71 3} 14. Bxf6 {+0.58/6 2.8} Qxf6 {+0.19/71 3} 15. Qe3 {+0.48/6 2.7} O-O {+0.30/71 3} 16. O-O {+0.12/6 2.6} Rfe8 {+0.47/71 3} 17. Nxe5 {-0.02/6 2.5} Bxg2 {+0.64/71 3} 18. Kxg2 {-0.26/6 2.4} Rxe5 {+0.62/71 3} 19. Qf3 {+0.50/6 2.3} Qe6 {+0.59/71 3} 20. Qb7 {+0.68/6 2.2} Re8 {+0.53/71 3} 21. e4 {+0.83/6 2.2} Rxe4 {+1.61/71 3} 22. Qxa7 {+0.92/6 2.1} d5 {+1.38/71 3} 23. cxd5 {+0.02/6 2.0} Qxd5 {+1.54/71 3} 24. Kg1 {-0.56/6 2.0} Qe6 {+1.15/71 3} 25. Rfd1 {-0.56/6 1.9} d5 {+1.19/71 3} 26. Rac1 {-0.44/5 1.8} d4 {+0.77/71 3} 27. Qa4 {-0.27/6 1.8} Rd8 {+1.48/71 3} 28. Qa6 {-0.56/5 1.7} Qxb3 {+2.09/71 3} 29. Rb1 {-0.66/5 1.7} Qe6 {+1.87/71 3} 30. Rxb6 {-0.60/5 1.6} Re1+ {+2.24/71 3} 31. Rxe1 {-0.75/6 1.5} Qxe1+ {+2.03/71 0.8} 32. Qf1 {-0.70/6 1.5} Qc3 {+1.58/71 3} 33. a4 {-0.73/6 1.4} d3 {+3.06/71 3} 34. Qd1 {-0.73/6 1.4} d2 {+6.19/71 3} 35. Rb1 {-2.58/6 1.4} Re8 {+9.09/71 4} 36. Kg2 {-3.78/5 1.3} Re1 {+10.24/71 4} 37. a5 {-7.42/6 1.3} Qc6+ {+14.02/71 4} 38. Kh3 {-7.42/6 1.2} Rxd1 {+13.28/71 4} 39. Rxd1 {-7.42/6 1.2} Qc1 {+15.05/71 4} 40. Rxc1 {-8.72/8 1.1} dxc1=Q {+15.98/71 1.1} 41. f4 {-8.81/7 5} Qf1+ {+15.56/71 2.9} 42. Kh4 {-9.00/7 5} Qe2 {+16.89/71 2.9} 43. Kh3 {-9.11/8 5} f5 {+99.96/71 2.9} 44. a6 {-99.94/7 0.7} g5 {+99.92/71 2.9} 45. fxg5 {-99.96/5 0.1} hxg5 {+99.96/71 3} 46. a7 {-99.98/3 0.1} Qf1# {+99.98/71 0.8} {Xboard adjudication: Checkmate} 0-1 [Event "Computer Chess Game"] [Site "dani-s-xp"] [Date "2018.09.20"] [Round "5"] [White "scorpio"] [Black "tscp"] [Result "1-0"] [TimeControl "40/120"] [Annotator "1. +0.00 9... -0.43"] 1. e4 {+0.00/72} c5 2. Nf3 {+0.00/72 0.1} Nc6 3. d4 {+0.00/72 0.1} cxd4 4. Nxd4 {+0.00/72 0.1} Nf6 5. Nc3 {+0.00/72 0.1} d6 6. Bg5 {+0.00/72 0.1} e6 7. Qd2 {+0.00/72 0.1} Be7 8. O-O-O {+0.00/72 0.1} O-O 9. f4 {+0.00/72 0.1} Nxd4 {-0.43/7 4} 10. Qxd4 {+0.00/72 0.1} Nh5 {-0.48/6 4} 11. Bxe7 {+1.30/71 7} Qxe7 {-0.50/7 4} 12. g3 {+1.36/71 4} e5 {-0.42/6 4} 13. Qxd6 {+2.10/71 4} Qxd6 {-0.12/7 3} 14. Rxd6 {+2.11/71 0.9} exf4 {-0.82/7 3} 15. Be2 {+2.37/71 4} Nf6 {-1.25/7 3} 16. gxf4 {+2.30/71 4} Re8 {-1.35/6 3} 17. Rg1 {+3.28/71 4} Re6 {-1.54/6 3} 18. Rd8+ {+3.52/71 4} Re8 {-1.49/6 2.9} 19. Rxe8+ {+3.20/71 4} Nxe8 {-1.39/6 2.8} 20. Rd1 {+2.86/71 4} Kf8 {-1.34/7 2.7} 21. b3 {+2.94/71 4} Bh3 {-1.21/6 2.6} 22. e5 {+2.70/71 4} Bg2 {-1.16/6 2.6} 23. Kb2 {+2.83/71 4} Ke7 {-0.98/6 2.5} 24. Bc4 {+2.49/71 4} Rc8 {-0.93/6 2.4} 25. Bd5 {+2.86/71 4} Bxd5 {-0.89/7 2.3} 26. Nxd5+ {+2.37/71 4} Ke6 {-1.02/6 2.2} 27. Rd4 {+2.49/71 4} f6 {-0.57/6 2.2} 28. exf6 {+2.12/71 4} Nxf6 {-0.87/7 2.1} 29. Nxf6 {+1.51/71 4} gxf6 {-0.88/7 2.0} 30. Re4+ {+1.64/71 4} Kf5 {-0.81/6 2.0} 31. Re7 {+1.72/71 4} Rb8 {-1.04/8 1.9} 32. Rd7 {+1.56/71 4} h6 {-0.67/7 1.8} 33. c4 {+1.01/71 4} Kxf4 {+0.00/7 1.8} 34. Rh7 {+0.93/71 4} Kg5 {-0.04/7 1.7} 35. h3 {+0.61/71 4} Kg6 {+0.27/7 1.6} 36. Rd7 {+0.17/71 4} f5 {+0.15/7 1.6} 37. Kc1 {+0.36/71 4} f4 {+0.34/7 1.5} 38. Rd3 {-0.28/71 4} Rf8 {+0.70/7 1.5} 39. Kd2 {-0.23/71 4} Rf6 {+0.59/6 1.4} 40. Ke2 {+0.01/71 5} Rf7 {+0.46/6 1.4} 41. c5 {+1.11/71 2.9} Kf5 {+0.70/7 5} 42. b4 {+0.28/71 2.9} Ke4 {+0.81/7 5} 43. Rd8 {-0.02/71 2.9} Re7 {+0.74/7 5} 44. Rg8 {+0.73/71 2.9} Kf5+ {+0.67/7 5} 45. Kf1 {+0.06/71 2.9} Re4 {+0.94/7 5} 46. Rb8 {+0.39/71 2.9} Rxb4 {+0.79/8 4} 47. c6 {+3.15/71 2.9} Rb1+ {-0.80/7 4} 48. Kf2 {+4.04/71 2.9} Rb2+ {-0.45/7 4} 49. Kf3 {+4.60/71 2.9} Rxa2 {-0.86/7 4} 50. Rf8+ {+4.25/71 0.7} Kg5 {-2.24/7 4} 51. cxb7 {+4.88/71 3} Ra3+ {-2.24/7 4} 52. Ke2 {+4.85/71 3} Ra2+ {-2.23/7 4} 53. Kd1 {+4.95/71 3} Rb2 {-2.35/8 4} 54. b8=Q {+4.78/71 3} Rxb8 {-2.61/9 3} 55. Rxb8 {+5.36/71 3} Kh4 {-2.82/8 3} 56. Rb3 {+5.56/71 3} Kg5 {-3.79/9 3} 57. Ke2 {+6.14/71 3} Kh4 {-4.03/8 3} 58. Ra3 {+7.24/71 3} Kg5 {-4.33/8 3} 59. Rxa7 {+7.55/71 3} Kh4 {-5.15/8 2.9} 60. Kf3 {+50.00/71 3} Kxh3 {-5.45/8 2.8} 61. Kxf4 {+53.84/71 0.8} Kg2 {-5.55/8 2.7} 62. Ra2+ {+54.94/71 3} Kg1 {-5.45/7 2.6} 63. Ra1+ {+53.76/71 3} Kg2 {-5.35/8 2.5} 64. Kg4 {+53.18/71 3} Kf2 {-5.35/9 2.5} 65. Ra6 {+55.51/71 3} Ke3 {-5.25/9 2.4} 66. Rxh6 {+55.91/71 0.8} Ke4 {-5.45/9 2.3} 67. Rh7 {+54.46/71 3} Ke5 {-5.45/9 2.2} 68. Rg7 {+54.54/71 3} Kf6 {-5.45/9 2.1} 69. Rg5 {+54.74/71 3} Ke6 {-5.45/9 2.1} 70. Ra5 {+54.64/71 3} Kd6 {-5.45/9 2.0} 71. Kf5 {+54.79/71 3} Kc6 {-5.55/8 1.9} 72. Ke5 {+54.84/71 3} Kb6 {-5.55/9 1.9} 73. Rd5 {+54.91/71 3} Kc6 {-5.55/9 1.8} 74. Kd4 {+99.92/71 3} Kb7 {-5.65/8 1.7} 75. Kc5 {+99.92/71 3} Kc7 {-5.75/8 1.7} 76. Rd4 {+99.94/71 3} Kb7 {-5.85/8 1.6} 77. Rd6 {+99.92/71 4} Kc7 {-5.85/9 1.6} 78. Rd3 {+99.94/71 4} Kb7 {-5.85/8 1.5} 79. Rd7+ {+99.92/71 4} Kc8 {-99.92/7 0.5} 80. Kc6 {+99.94/71 4} Kb8 {-99.94/6 0.1} 81. Re7 {+99.96/71 2.9} Ka8 {-99.96/5 0.1} 82. Kb6 {+99.96/71 2.9} Kb8 {-99.98/3 0.1} 83. Re8# {+99.98/71 2.9} {Xboard adjudication: Checkmate} 1-0 [Event "Computer Chess Game"] [Site "dani-s-xp"] [Date "2018.09.20"] [Round "6"] [White "tscp"] [Black "scorpio"] [Result "0-1"] [TimeControl "40/120"] [Annotator "7. -0.20 1... +0.00"] 1. d4 Nf6 {+0.00/72 0.1} 2. c4 e6 {+0.00/72 0.1} 3. Nf3 b6 {+0.00/72 0.1} 4. g3 Ba6 {+0.00/72 0.1} 5. b3 Bb4+ {+0.00/72 0.1} 6. Bd2 Be7 {+0.00/72 0.1} 7. Nc3 {-0.20/6 4} O-O {+0.00/72 0.1} 8. e4 {+0.18/6 4} d5 {+0.00/72 0.1} 9. e5 {+0.11/6 4} Ne4 {-0.06/71 7} 10. Bd3 {+0.21/6 4} Bb7 {-0.33/71 3} 11. Qc2 {+0.30/6 3} Nxd2 {+0.66/71 3} 12. Bxh7+ {+0.21/6 3} Kh8 {+0.00/72 0.1} 13. Nxd2 {+0.21/6 3} Bb4 {+2.00/71 3} 14. Ne2 {+0.01/6 3} g6 {+0.84/71 3} 15. Bxg6 {+0.30/6 3} fxg6 {+1.41/70 0.9} 16. Qxg6 {+1.23/6 2.9} Qe7 {+1.23/71 4} 17. Qh6+ {+1.19/6 2.8} Qh7 {+2.32/71 0.9} 18. Qxe6 {+0.64/6 2.7} dxc4 {+3.62/71 4} 19. O-O {-0.06/6 2.7} Bxd2 {+5.76/71 1.0} 20. Qxc4 {-1.22/5 2.6} Bf3 {+6.61/71 4} 21. h4 {+0.00/6 2.5} Qf5 {+7.40/71 4} 22. Kh2 {-0.44/5 2.4} c6 {+6.70/71 4} 23. Ng1 {-0.88/6 2.3} b5 {+6.21/71 4} 24. Qc5 {-0.65/6 2.2} Na6 {+6.56/71 4} 25. Qa3 {-0.35/6 2.2} Nb4 {+6.08/71 4} 26. Nxf3 {-0.55/6 2.1} Qxf3 {+7.33/71 1.0} 27. Rad1 {-0.97/6 2.0} Qe2 {+7.59/71 4} 28. Qb2 {-1.30/6 2.0} Rxf2+ {+14.22/71 1.1} 29. Rxf2 {-5.52/6 1.9} Qxf2+ {+14.87/71 1.1} 30. Kh1 {-6.89/6 1.8} Qf3+ {+15.41/71 1.2} 31. Kh2 {-7.24/6 1.8} Qe2+ {+15.18/71 1.3} 32. Kh3 {-7.60/6 1.7} Qxd1 {+16.37/71 1.4} 33. Kh2 {-8.43/6 1.6} Rf8 {+25.87/71 6} 34. Qxd2 {-15.50/6 1.6} Qxd2+ {+28.50/71 1.5} 35. Kh3 {-99.94/6 1.4} Qf2 {+99.78/71 7} 36. h5 {-99.96/5 0.1} Rf3 {+99.96/71 7} 37. e6 {-99.98/3 0.1} Qxg3# {+99.98/71 1.8} {Xboard adjudication: Checkmate} 0-1 [/copy][/pgn]
BrendanJNorman
Posts: 2526
Joined: Mon Feb 08, 2016 12:43 am
Full name: Brendan J Norman

Re: Scorpio MCTS+NN: What are the strongest nets?

Post by BrendanJNorman »

Daniel Shawul wrote: Fri Sep 21, 2018 6:35 am It seems that I have made a mistake in counting the number of blocks , so what I thought was a 2x32 net was actually a 3x32 net...same for other nets.
I retrained an actual 2x32 net in 10 minutes over 25000 games and recompiled then tensorflow with optimizations on.
Then this version has the same nps number as the the scorpio-mcts version without neural networks !!!
Also I am using min-max backup with mcts instead of the averaging as in the standard mcts. This has made it tactically very good
and easily destroys tscp as the first 6 games below show. Note that I am only talking about the CPU. I don't think
you will be able to beat this version unless you are >2400.

To answer your question, yes training new networks I am sure will be very fun. Also since my input planes are attack tables,
and I have separate nodes for piece counts, you can capture piece values and attack features very fast. Try it and let me know.

[pgn] [Event "Computer Chess Game"] [Site "dani-s-xp"] [Date "2018.09.20"] [Round "1"] [White "scorpio"] [Black "tscp"] [Result "1-0"] [TimeControl "40/120"] [Annotator "19. +1.90 12... -0.05"] 1. e4 e5 2. Nf3 Nc6 3. Bb5 a6 4. Ba4 Nf6 5. O-O Be7 6. Re1 b5 7. Bb3 O-O 8. c3 d6 9. h3 Na5 10. Bc2 c5 11. d4 Qc7 12. Nbd2 cxd4 {-0.05/6 4} 13. cxd4 Bd7 {+0.00/6 4} 14. Nf1 Rfc8 {-0.01/6 4} 15. Ne3 Nc6 {+0.05/6 4} 16. Nd5 Nxd5 {+0.18/6 3} 17. exd5 Nb4 {-0.06/6 3} 18. Bb1 f6 {+0.13/6 3} 19. dxe5 {+1.90/71 10} fxe5 {+0.02/6 3} 20. Bg5 {+0.64/71 5} Bxg5 {-0.04/6 3} 21. Nxg5 {+0.43/71 1.2} Nc2 {-0.09/6 2.9} 22. Ne6 {+0.72/71 5} Bxe6 {-0.06/6 2.8} 23. Bxc2 {+0.67/71 5} Bd7 {-0.19/6 2.7} 24. Rc1 {+0.80/71 5} Qa7 {-0.26/6 2.6} 25. Qh5 {+2.41/71 5} h6 {-0.49/6 2.6} 26. Qg6 {+1.35/71 5} Qc5 {-0.59/5 2.5} 27. Qh7+ {+6.27/71 5} Kf8 {-0.43/5 2.4} 28. Re3 {+6.87/71 5} Qxd5 {-0.12/5 2.3} 29. Rf3+ {+6.11/71 5} Qxf3 {-1.40/5 2.2} 30. gxf3 {+7.21/71 1.3} Bxh3 {-1.69/6 2.2} 31. Qg6 {+7.23/71 5} Rd8 {-1.60/6 2.1} 32. Bb3 {+10.59/71 6} d5 {-2.63/6 2.0} 33. Rc7 {+13.11/71 6} Rd7 {-3.13/6 1.9} 34. Rxd7 {+14.78/71 1.4} Bxd7 {-7.08/7 1.9} 35. Bxd5 {+16.21/71 6} Ke7 {-8.27/6 1.8} 36. Qxg7+ {+16.58/71 6} Kd6 {-9.29/6 1.8} 37. Bxa8 {+16.73/71 1.6} h5 {-9.76/7 1.7} 38. Qf8+ {+18.28/71 8} Ke6 {-10.65/6 1.6} 39. Be4 {+20.52/71 8} a5 {-11.43/6 1.6} 40. Bf5+ {+20.65/71 2.3} Kd5 {-12.25/6 1.5} 41. Bxd7 {+20.55/71 3} b4 {-13.10/7 5} 42. Qa8+ {+22.28/71 3} Kd6 {-13.93/6 5} 43. Qc6+ {+22.74/71 3} Ke7 {-14.73/6 5} 44. Qe6+ {+22.72/71 3} Kf8 {-14.77/6 5} 45. Qf6+ {+23.03/71 3} Kg8 {-99.92/6 1.7} 46. Qe7 {+99.75/71 3} h4 {-99.94/5 0.1} 47. Be6+ {+99.84/71 3} Kh8 {-99.96/3 0.1} 48. Qf8+ {+99.96/71 0.8} Kh7 {-99.98/2 0.1} 49. Bf5# {+99.98/71 0.8} {Xboard adjudication: Checkmate} 1-0 [Event "Computer Chess Game"] [Site "dani-s-xp"] [Date "2018.09.20"] [Round "2"] [White "tscp"] [Black "scorpio"] [Result "0-1"] [TimeControl "40/120"] [Annotator "5. +0.20 1... +0.00"] 1. Nf3 Nf6 {+0.00/72 0.1} 2. d4 e6 {+0.00/72 0.1} 3. c4 b6 {+0.00/72 0.1} 4. Nc3 Bb4 {+0.00/72 0.1} 5. Bd2 {+0.20/6 4} O-O {+0.00/72 0.1} 6. e3 {+0.25/7 4} Bb7 {+0.00/72 0.1} 7. Bd3 {+0.19/6 4} c5 {+0.00/72 0.1} 8. dxc5 {+0.26/6 4} Bxc5 {+0.13/71 7} 9. O-O {+0.23/6 3} d6 {+0.04/71 3} 10. e4 {+0.31/6 3} Qd7 {+0.33/71 3} 11. Bg5 {+0.62/5 3} Nc6 {-0.09/71 3} 12. Bxf6 {+0.89/5 3} gxf6 {-0.08/71 0.8} 13. Qd2 {+0.84/5 3} Qd8 {-0.03/71 3} 14. Qf4 {+0.94/5 2.9} a5 {+0.03/71 3} 15. Rad1 {+0.94/5 2.8} Qe7 {-0.22/71 3} 16. Rd2 {+1.02/5 2.7} Rad8 {-0.36/71 3} 17. Qg4+ {+1.04/5 2.7} Kh8 {+0.00/72 0.1} 18. Rfd1 {+1.02/5 2.6} Ne5 {+0.93/71 4} 19. Qh3 {+0.93/6 2.5} Ba8 {+0.31/71 4} 20. Nxe5 {+1.04/5 2.4} dxe5 {+0.74/71 4} 21. Qg4 {+0.89/5 2.3} Rg8 {+0.64/71 4} 22. Qh4 {+0.78/5 2.2} Rdf8 {+0.71/71 4} 23. Bb1 {+0.94/5 2.2} Bc6 {+0.45/71 4} 24. Qh3 {+0.79/5 2.1} Re8 {+1.21/71 4} 25. Bd3 {+0.80/5 2.0} Rb8 {+1.01/71 4} 26. Bc2 {+0.84/4 2.0} f5 {+0.52/71 4} 27. Re2 {+0.76/4 1.9} f4 {+1.01/71 4} 28. Red2 {+0.74/5 1.8} Rbc8 {+1.08/71 4} 29. Ba4 {+0.76/4 1.8} Bxa4 {+1.04/71 4} 30. Nxa4 {+0.56/6 1.7} Bd4 {+1.23/71 4} 31. b3 {+0.51/5 1.7} Qf6 {+1.28/71 4} 32. Qf3 {+0.58/5 1.6} Rb8 {+1.61/71 4} 33. Nc3 {+0.68/5 1.5} Bc5 {+0.58/71 4} 34. Rd7 {+0.80/5 1.5} Rbf8 {+0.20/71 4} 35. Rb7 {+0.97/4 1.4} Qh4 {+0.47/71 4} 36. Na4 {+1.73/5 1.4} Bd4 {+0.53/71 4} 37. Nxb6 {+2.20/5 1.3} Rd8 {+0.23/71 4} 38. Rc1 {+2.12/5 1.3} f5 {-0.71/71 4} 39. Rc2 {+2.15/5 1.3} Bxb6 {+4.35/71 4} 40. exf5 {-0.84/6 1.2} Qh3 {+10.01/71 4} 41. Kf1 {-2.14/7 5} Qxf5 {+4.98/71 2.9} 42. Qe2 {-1.27/6 5} Rxg2 {+8.71/71 2.9} 43. Kxg2 {-2.35/6 5} f3+ {+10.34/71 0.7} 44. Qxf3 {-4.20/7 5} Rg8+ {+10.51/71 0.7} 45. Qg3 {-4.40/7 4} Rxg3+ {+10.80/71 0.8} 46. hxg3 {-4.40/6 4} Qe4+ {+11.18/71 0.8} 47. Kg1 {-4.70/7 4} Qxb7 {+10.49/71 0.8} 48. Rd2 {-5.55/7 4} Qf3 {+12.15/71 3} 49. c5 {-7.15/7 4} Bxc5 {+13.19/71 0.8} 50. g4 {-7.78/7 4} Bb6 {+14.72/71 3} 51. g5 {-8.15/7 4} Qg3+ {+14.93/71 3} 52. Kf1 {-7.53/6 4} Qh3+ {+14.71/71 0.8} 53. Ke1 {-7.75/6 3} Qh1+ {+15.30/71 3} 54. Ke2 {-7.95/6 3} Qe4+ {+14.37/71 3} 55. Kf1 {-7.42/6 3} Bd4 {+13.77/71 3} 56. a3 {-8.20/6 3} Qh1+ {+15.25/71 3} 57. Ke2 {-8.27/6 3} Qh5+ {+14.83/71 3} 58. f3 {-8.18/6 2.9} Qxg5 {+15.08/71 0.9} 59. Rd3 {-8.20/7 2.8} Qg2+ {+15.37/71 3} 60. Kd1 {-8.87/6 2.7} Qg1+ {+15.51/71 3} 61. Kd2 {-8.97/6 2.6} Qf2+ {+15.35/71 3} 62. Kd1 {-9.02/6 2.5} Qb2 {+15.40/71 4} 63. b4 {-8.74/7 2.5} Qb1+ {+17.13/71 4} 64. Ke2 {-8.62/6 2.4} Qa2+ {+16.05/71 4} 65. Kd1 {-8.64/6 2.3} Qc4 {+15.47/71 4} 66. Kd2 {-8.73/7 2.2} axb4 {+16.07/71 4} 67. axb4 {-8.93/7 2.1} Kg7 {+15.63/71 4} 68. b5 {-9.13/7 2.1} Qxb5 {+16.71/71 4} 69. Kc2 {-9.20/6 2.0} Qb2+ {+17.31/71 4} 70. Kd1 {-10.28/5 1.9} Kf6 {+17.24/71 4} 71. Rd2 {-10.95/7 1.9} Qb3+ {+18.50/71 4} 72. Ke2 {-10.95/5 1.8} Qc4+ {+17.81/71 4} 73. Kd1 {-10.95/5 1.8} Qf1+ {+17.39/71 4} 74. Kc2 {-10.95/5 1.7} Ke7 {+17.06/71 4} 75. Rd3 {-10.90/6 1.6} Qe2+ {+17.79/71 4} 76. Rd2 {-11.00/6 1.6} Qxf3 {+18.23/71 4} 77. Rd3 {-11.09/6 1.5} Qe2+ {+18.55/71 4} 78. Rd2 {-11.09/5 1.5} Qc4+ {+19.79/71 4} 79. Kd1 {-12.38/5 1.4} Be3 {+21.22/71 4} 80. Rh2 {-12.80/6 1.4} Qd3+ {+23.70/71 1.1} 81. Ke1 {-12.70/6 5} Bf4 {+21.14/71 2.9} 82. Rg2 {-13.14/6 5} Qe3+ {+22.33/71 0.7} 83. Kf1 {-12.80/6 5} Bg3 {+20.52/71 3} 84. Ra2 {-12.72/7 5} Qf3+ {+20.23/71 0.8} 85. Kg1 {-12.72/6 5} Kf6 {+18.17/71 0.8} 86. Rb2 {-12.90/7 4} Bf2+ {+27.65/71 3} 87. Rxf2 {-5.28/10 4} Qxf2+ {+51.74/71 3} 88. Kxf2 {-4.96/12 4} h5 {+53.19/71 3} 89. Kg3 {-4.93/11 4} e4 {+52.92/71 3} 90. Kf4 {-3.66/12 4} h4 {+52.66/71 3} 91. Kxe4 {-3.66/11 4} h3 {+53.73/71 3} 92. Kf3 {-2.06/11 4} h2 {+54.04/71 3} 93. Kg2 {-2.06/12 4} h1=Q+ {+57.36/71 0.8} 94. Kxh1 {-2.06/11 3} Ke5 {+54.11/71 3} 95. Kg2 {-2.30/13 3} Kf4 {+54.51/71 3} 96. Kf2 {-2.16/13 3} e5 {+54.91/71 3} 97. Ke2 {-2.16/12 3} Ke4 {+0.00/72 0.1} 98. Kd2 {-8.70/12 3} Kf3 {+54.18/71 3} 99. Ke1 {-8.80/12 2.9} e4 {+55.06/71 3} 100. Kd1 {-8.90/12 2.8} e3 {+55.12/71 3} 101. Ke1 {-9.10/11 2.7} e2 {+0.00/72 0.1} 102. Kd2 {-9.20/11 2.6} Kf2 {+0.00/72 0.1} 103. Kd3 {-9.20/9 2.5} e1=Q {+57.86/71 1.0} 104. Kd4 {-9.20/7 2.4} Qe2 {+56.41/71 4} 105. Kd5 {-9.20/7 2.4} Ke3 {+56.56/71 4} 106. Ke5 {-9.30/7 2.3} Qg2 {+56.69/71 4} 107. Kd6 {-9.30/7 2.2} Kd4 {+56.71/71 4} 108. Ke6 {-9.40/7 2.1} Qd5+ {+99.92/71 4} 109. Kf6 {-9.40/6 2.1} Qe4 {+99.92/71 4} 110. Kf7 {-9.40/7 2.0} Ke5 {+99.92/71 4} 111. Ke8 {-9.40/6 1.9} Qh7 {+99.94/71 4} 112. Kd8 {-99.96/5 0.1} Kd6 {+99.96/71 4} 113. Ke8 {-99.98/3 0.1} Qg8# {+99.98/71 4} {Xboard adjudication: Checkmate} 0-1 [Event "Computer Chess Game"] [Site "dani-s-xp"] [Date "2018.09.20"] [Round "3"] [White "scorpio"] [Black "tscp"] [Result "1-0"] [TimeControl "40/120"] [Annotator "1. +0.00 6... -0.59"] 1. e4 {+0.00/72} c5 2. Nf3 {+0.00/72 0.1} e6 3. d4 {+0.00/72 0.1} cxd4 4. Nxd4 {+0.00/72 0.1} Nc6 5. Nc3 {+0.00/72 0.1} Qc7 6. Be3 {+0.00/72 0.1} Bb4 {-0.59/6 4} 7. Nb5 {+1.57/71 7} Bxc3+ {-0.63/6 4} 8. bxc3 {+3.12/71 3} Qb8 {-0.52/6 4} 9. Nd6+ {+3.15/71 3} Kf8 {-0.75/5 4} 10. Rb1 {+3.28/71 3} Nge7 {-0.55/6 3} 11. Bf4 {+3.22/71 3} e5 {-0.65/6 3} 12. Bc1 {+2.70/71 3} Kg8 {-0.26/6 3} 13. Bc4 {+4.06/71 3} Nd8 {-0.76/6 3} 14. Bg5 {+3.86/71 3} Kf8 {-0.69/6 3} 15. f4 {+4.42/71 3} exf4 {-0.90/6 2.9} 16. Rf1 {+4.29/71 3} Ne6 {-0.61/6 2.8} 17. Bxf4 {+4.04/71 3} Nxf4 {-0.52/6 2.7} 18. Rxf4 {+5.45/71 0.8} f6 {-1.15/6 2.7} 19. Qd4 {+7.76/71 3} Nf5 {-3.69/6 2.6} 20. Rxf5 {+10.51/71 0.9} Qxd6 {-10.71/6 2.5} 21. Qxd6+ {+19.25/71 0.9} Ke8 {-10.71/5 2.4} 22. e5 {+19.34/71 4} Kd8 {-10.94/6 2.3} 23. exf6 {+19.89/71 4} Re8+ {-14.95/6 2.2} 24. Kd2 {+19.97/71 4} gxf6 {-16.60/6 2.2} 25. Qxf6+ {+25.46/71 4} Re7 {-17.32/5 2.1} 26. Re1 {+27.55/71 4} b6 {-17.53/6 2.0} 27. Qd6 {+99.69/71 4} Rxe1 {-99.96/5 0.2} 28. Kxe1 {+99.91/71 4} Bb7 {-99.98/3 0.1} 29. Rf8# {+99.98/71 0.9} {Xboard adjudication: Checkmate} 1-0 [Event "Computer Chess Game"] [Site "dani-s-xp"] [Date "2018.09.20"] [Round "4"] [White "tscp"] [Black "scorpio"] [Result "0-1"] [TimeControl "40/120"] [Annotator "3. -0.78 1... +0.00"] 1. c4 e5 {+0.00/72 0.1} 2. g3 Nf6 {+0.00/72 0.1} 3. Nc3 {-0.78/6 4} Bb4 {+0.00/72 0.1} 4. a3 {-0.69/6 4} Bd6 {+0.26/71 6} 5. Nf3 {-0.46/6 4} b6 {-0.21/71 3} 6. d4 {-0.11/6 4} exd4 {+0.00/71 3} 7. Nb5 {-0.01/6 3} Bb7 {-0.15/71 3} 8. Nxd6+ {+0.23/6 3} cxd6 {-0.22/71 0.8} 9. Qxd4 {+0.15/6 3} Qe7 {-0.14/71 3} 10. Qc3 {+0.23/6 3} Nc6 {-0.08/71 3} 11. Bg5 {+0.41/5 3} Ne5 {+0.22/71 3} 12. Bg2 {+0.41/6 2.9} Rc8 {+0.52/71 3} 13. b3 {+0.37/6 2.8} h6 {+0.26/71 3} 14. Bxf6 {+0.58/6 2.8} Qxf6 {+0.19/71 3} 15. Qe3 {+0.48/6 2.7} O-O {+0.30/71 3} 16. O-O {+0.12/6 2.6} Rfe8 {+0.47/71 3} 17. Nxe5 {-0.02/6 2.5} Bxg2 {+0.64/71 3} 18. Kxg2 {-0.26/6 2.4} Rxe5 {+0.62/71 3} 19. Qf3 {+0.50/6 2.3} Qe6 {+0.59/71 3} 20. Qb7 {+0.68/6 2.2} Re8 {+0.53/71 3} 21. e4 {+0.83/6 2.2} Rxe4 {+1.61/71 3} 22. Qxa7 {+0.92/6 2.1} d5 {+1.38/71 3} 23. cxd5 {+0.02/6 2.0} Qxd5 {+1.54/71 3} 24. Kg1 {-0.56/6 2.0} Qe6 {+1.15/71 3} 25. Rfd1 {-0.56/6 1.9} d5 {+1.19/71 3} 26. Rac1 {-0.44/5 1.8} d4 {+0.77/71 3} 27. Qa4 {-0.27/6 1.8} Rd8 {+1.48/71 3} 28. Qa6 {-0.56/5 1.7} Qxb3 {+2.09/71 3} 29. Rb1 {-0.66/5 1.7} Qe6 {+1.87/71 3} 30. Rxb6 {-0.60/5 1.6} Re1+ {+2.24/71 3} 31. Rxe1 {-0.75/6 1.5} Qxe1+ {+2.03/71 0.8} 32. Qf1 {-0.70/6 1.5} Qc3 {+1.58/71 3} 33. a4 {-0.73/6 1.4} d3 {+3.06/71 3} 34. Qd1 {-0.73/6 1.4} d2 {+6.19/71 3} 35. Rb1 {-2.58/6 1.4} Re8 {+9.09/71 4} 36. Kg2 {-3.78/5 1.3} Re1 {+10.24/71 4} 37. a5 {-7.42/6 1.3} Qc6+ {+14.02/71 4} 38. Kh3 {-7.42/6 1.2} Rxd1 {+13.28/71 4} 39. Rxd1 {-7.42/6 1.2} Qc1 {+15.05/71 4} 40. Rxc1 {-8.72/8 1.1} dxc1=Q {+15.98/71 1.1} 41. f4 {-8.81/7 5} Qf1+ {+15.56/71 2.9} 42. Kh4 {-9.00/7 5} Qe2 {+16.89/71 2.9} 43. Kh3 {-9.11/8 5} f5 {+99.96/71 2.9} 44. a6 {-99.94/7 0.7} g5 {+99.92/71 2.9} 45. fxg5 {-99.96/5 0.1} hxg5 {+99.96/71 3} 46. a7 {-99.98/3 0.1} Qf1# {+99.98/71 0.8} {Xboard adjudication: Checkmate} 0-1 [Event "Computer Chess Game"] [Site "dani-s-xp"] [Date "2018.09.20"] [Round "5"] [White "scorpio"] [Black "tscp"] [Result "1-0"] [TimeControl "40/120"] [Annotator "1. +0.00 9... -0.43"] 1. e4 {+0.00/72} c5 2. Nf3 {+0.00/72 0.1} Nc6 3. d4 {+0.00/72 0.1} cxd4 4. Nxd4 {+0.00/72 0.1} Nf6 5. Nc3 {+0.00/72 0.1} d6 6. Bg5 {+0.00/72 0.1} e6 7. Qd2 {+0.00/72 0.1} Be7 8. O-O-O {+0.00/72 0.1} O-O 9. f4 {+0.00/72 0.1} Nxd4 {-0.43/7 4} 10. Qxd4 {+0.00/72 0.1} Nh5 {-0.48/6 4} 11. Bxe7 {+1.30/71 7} Qxe7 {-0.50/7 4} 12. g3 {+1.36/71 4} e5 {-0.42/6 4} 13. Qxd6 {+2.10/71 4} Qxd6 {-0.12/7 3} 14. Rxd6 {+2.11/71 0.9} exf4 {-0.82/7 3} 15. Be2 {+2.37/71 4} Nf6 {-1.25/7 3} 16. gxf4 {+2.30/71 4} Re8 {-1.35/6 3} 17. Rg1 {+3.28/71 4} Re6 {-1.54/6 3} 18. Rd8+ {+3.52/71 4} Re8 {-1.49/6 2.9} 19. Rxe8+ {+3.20/71 4} Nxe8 {-1.39/6 2.8} 20. Rd1 {+2.86/71 4} Kf8 {-1.34/7 2.7} 21. b3 {+2.94/71 4} Bh3 {-1.21/6 2.6} 22. e5 {+2.70/71 4} Bg2 {-1.16/6 2.6} 23. Kb2 {+2.83/71 4} Ke7 {-0.98/6 2.5} 24. Bc4 {+2.49/71 4} Rc8 {-0.93/6 2.4} 25. Bd5 {+2.86/71 4} Bxd5 {-0.89/7 2.3} 26. Nxd5+ {+2.37/71 4} Ke6 {-1.02/6 2.2} 27. Rd4 {+2.49/71 4} f6 {-0.57/6 2.2} 28. exf6 {+2.12/71 4} Nxf6 {-0.87/7 2.1} 29. Nxf6 {+1.51/71 4} gxf6 {-0.88/7 2.0} 30. Re4+ {+1.64/71 4} Kf5 {-0.81/6 2.0} 31. Re7 {+1.72/71 4} Rb8 {-1.04/8 1.9} 32. Rd7 {+1.56/71 4} h6 {-0.67/7 1.8} 33. c4 {+1.01/71 4} Kxf4 {+0.00/7 1.8} 34. Rh7 {+0.93/71 4} Kg5 {-0.04/7 1.7} 35. h3 {+0.61/71 4} Kg6 {+0.27/7 1.6} 36. Rd7 {+0.17/71 4} f5 {+0.15/7 1.6} 37. Kc1 {+0.36/71 4} f4 {+0.34/7 1.5} 38. Rd3 {-0.28/71 4} Rf8 {+0.70/7 1.5} 39. Kd2 {-0.23/71 4} Rf6 {+0.59/6 1.4} 40. Ke2 {+0.01/71 5} Rf7 {+0.46/6 1.4} 41. c5 {+1.11/71 2.9} Kf5 {+0.70/7 5} 42. b4 {+0.28/71 2.9} Ke4 {+0.81/7 5} 43. Rd8 {-0.02/71 2.9} Re7 {+0.74/7 5} 44. Rg8 {+0.73/71 2.9} Kf5+ {+0.67/7 5} 45. Kf1 {+0.06/71 2.9} Re4 {+0.94/7 5} 46. Rb8 {+0.39/71 2.9} Rxb4 {+0.79/8 4} 47. c6 {+3.15/71 2.9} Rb1+ {-0.80/7 4} 48. Kf2 {+4.04/71 2.9} Rb2+ {-0.45/7 4} 49. Kf3 {+4.60/71 2.9} Rxa2 {-0.86/7 4} 50. Rf8+ {+4.25/71 0.7} Kg5 {-2.24/7 4} 51. cxb7 {+4.88/71 3} Ra3+ {-2.24/7 4} 52. Ke2 {+4.85/71 3} Ra2+ {-2.23/7 4} 53. Kd1 {+4.95/71 3} Rb2 {-2.35/8 4} 54. b8=Q {+4.78/71 3} Rxb8 {-2.61/9 3} 55. Rxb8 {+5.36/71 3} Kh4 {-2.82/8 3} 56. Rb3 {+5.56/71 3} Kg5 {-3.79/9 3} 57. Ke2 {+6.14/71 3} Kh4 {-4.03/8 3} 58. Ra3 {+7.24/71 3} Kg5 {-4.33/8 3} 59. Rxa7 {+7.55/71 3} Kh4 {-5.15/8 2.9} 60. Kf3 {+50.00/71 3} Kxh3 {-5.45/8 2.8} 61. Kxf4 {+53.84/71 0.8} Kg2 {-5.55/8 2.7} 62. Ra2+ {+54.94/71 3} Kg1 {-5.45/7 2.6} 63. Ra1+ {+53.76/71 3} Kg2 {-5.35/8 2.5} 64. Kg4 {+53.18/71 3} Kf2 {-5.35/9 2.5} 65. Ra6 {+55.51/71 3} Ke3 {-5.25/9 2.4} 66. Rxh6 {+55.91/71 0.8} Ke4 {-5.45/9 2.3} 67. Rh7 {+54.46/71 3} Ke5 {-5.45/9 2.2} 68. Rg7 {+54.54/71 3} Kf6 {-5.45/9 2.1} 69. Rg5 {+54.74/71 3} Ke6 {-5.45/9 2.1} 70. Ra5 {+54.64/71 3} Kd6 {-5.45/9 2.0} 71. Kf5 {+54.79/71 3} Kc6 {-5.55/8 1.9} 72. Ke5 {+54.84/71 3} Kb6 {-5.55/9 1.9} 73. Rd5 {+54.91/71 3} Kc6 {-5.55/9 1.8} 74. Kd4 {+99.92/71 3} Kb7 {-5.65/8 1.7} 75. Kc5 {+99.92/71 3} Kc7 {-5.75/8 1.7} 76. Rd4 {+99.94/71 3} Kb7 {-5.85/8 1.6} 77. Rd6 {+99.92/71 4} Kc7 {-5.85/9 1.6} 78. Rd3 {+99.94/71 4} Kb7 {-5.85/8 1.5} 79. Rd7+ {+99.92/71 4} Kc8 {-99.92/7 0.5} 80. Kc6 {+99.94/71 4} Kb8 {-99.94/6 0.1} 81. Re7 {+99.96/71 2.9} Ka8 {-99.96/5 0.1} 82. Kb6 {+99.96/71 2.9} Kb8 {-99.98/3 0.1} 83. Re8# {+99.98/71 2.9} {Xboard adjudication: Checkmate} 1-0 [Event "Computer Chess Game"] [Site "dani-s-xp"] [Date "2018.09.20"] [Round "6"] [White "tscp"] [Black "scorpio"] [Result "0-1"] [TimeControl "40/120"] [Annotator "7. -0.20 1... +0.00"] 1. d4 Nf6 {+0.00/72 0.1} 2. c4 e6 {+0.00/72 0.1} 3. Nf3 b6 {+0.00/72 0.1} 4. g3 Ba6 {+0.00/72 0.1} 5. b3 Bb4+ {+0.00/72 0.1} 6. Bd2 Be7 {+0.00/72 0.1} 7. Nc3 {-0.20/6 4} O-O {+0.00/72 0.1} 8. e4 {+0.18/6 4} d5 {+0.00/72 0.1} 9. e5 {+0.11/6 4} Ne4 {-0.06/71 7} 10. Bd3 {+0.21/6 4} Bb7 {-0.33/71 3} 11. Qc2 {+0.30/6 3} Nxd2 {+0.66/71 3} 12. Bxh7+ {+0.21/6 3} Kh8 {+0.00/72 0.1} 13. Nxd2 {+0.21/6 3} Bb4 {+2.00/71 3} 14. Ne2 {+0.01/6 3} g6 {+0.84/71 3} 15. Bxg6 {+0.30/6 3} fxg6 {+1.41/70 0.9} 16. Qxg6 {+1.23/6 2.9} Qe7 {+1.23/71 4} 17. Qh6+ {+1.19/6 2.8} Qh7 {+2.32/71 0.9} 18. Qxe6 {+0.64/6 2.7} dxc4 {+3.62/71 4} 19. O-O {-0.06/6 2.7} Bxd2 {+5.76/71 1.0} 20. Qxc4 {-1.22/5 2.6} Bf3 {+6.61/71 4} 21. h4 {+0.00/6 2.5} Qf5 {+7.40/71 4} 22. Kh2 {-0.44/5 2.4} c6 {+6.70/71 4} 23. Ng1 {-0.88/6 2.3} b5 {+6.21/71 4} 24. Qc5 {-0.65/6 2.2} Na6 {+6.56/71 4} 25. Qa3 {-0.35/6 2.2} Nb4 {+6.08/71 4} 26. Nxf3 {-0.55/6 2.1} Qxf3 {+7.33/71 1.0} 27. Rad1 {-0.97/6 2.0} Qe2 {+7.59/71 4} 28. Qb2 {-1.30/6 2.0} Rxf2+ {+14.22/71 1.1} 29. Rxf2 {-5.52/6 1.9} Qxf2+ {+14.87/71 1.1} 30. Kh1 {-6.89/6 1.8} Qf3+ {+15.41/71 1.2} 31. Kh2 {-7.24/6 1.8} Qe2+ {+15.18/71 1.3} 32. Kh3 {-7.60/6 1.7} Qxd1 {+16.37/71 1.4} 33. Kh2 {-8.43/6 1.6} Rf8 {+25.87/71 6} 34. Qxd2 {-15.50/6 1.6} Qxd2+ {+28.50/71 1.5} 35. Kh3 {-99.94/6 1.4} Qf2 {+99.78/71 7} 36. h5 {-99.96/5 0.1} Rf3 {+99.96/71 7} 37. e6 {-99.98/3 0.1} :? Qxg3# {+99.98/71 1.8} {Xboard adjudication: Checkmate} 0-1 [/copy][/pgn]
Hi Daniel,

This new config looks great.

Dying to give it a run against myself (and my Rodent personalities!)

Where can I find the new download? :)
User avatar
Guenther
Posts: 4606
Joined: Wed Oct 01, 2008 6:33 am
Location: Regensburg, Germany
Full name: Guenther Simon

Re: Scorpio MCTS+NN: What are the strongest nets?

Post by Guenther »

Daniel, I don't understand your output from those 6 games.
It looks as if 0.00/72 0.1 is the Scorpio MCTS output for being
still in book?
But why is there no output then in the first game with the huge
(for my taste too long) opening line?
Daniel Shawul wrote: Fri Sep 21, 2018 6:35 am [Annotator "19. +1.90 12... -0.05"]

1. e4 e5 2. Nf3 Nc6 3. Bb5 a6 4. Ba4 Nf6 5. O-O Be7 6. Re1 b5 7. Bb3 O-O 8.
c3 d6 9. h3 Na5 10. Bc2 c5 11. d4 Qc7 12. Nbd2 cxd4 {-0.05/6 4} 13. cxd4
Bd7 {+0.00/6 4} 14. Nf1 Rfc8 {-0.01/6 4} 15. Ne3 Nc6 {+0.05/6 4} 16. Nd5
Nxd5 {+0.18/6 3} 17. exd5 Nb4 {-0.06/6 3} 18. Bb1 f6 {+0.13/6 3} 19. dxe5
{+1.90/71 10}

[Annotator "5. +0.20 1... +0.00"]

1. Nf3 Nf6 {+0.00/72 0.1} 2. d4 e6 {+0.00/72 0.1} 3. c4 b6 {+0.00/72 0.1}
4. Nc3 Bb4 {+0.00/72 0.1} 5. Bd2 {+0.20/6 4} O-O {+0.00/72 0.1}

[Annotator "1. +0.00 6... -0.59"]

1. e4 {+0.00/72} c5 2. Nf3 {+0.00/72 0.1} e6 3. d4 {+0.00/72 0.1} cxd4 4.
Nxd4 {+0.00/72 0.1} Nc6 5. Nc3 {+0.00/72 0.1} Qc7 6. Be3 {+0.00/72 0.1} Bb4
{-0.59/6 4}

[Annotator "3. -0.78 1... +0.00"]

1. c4 e5 {+0.00/72 0.1} 2. g3 Nf6 {+0.00/72 0.1} 3. Nc3 {-0.78/6 4} Bb4
{+0.00/72 0.1}

[Annotator "1. +0.00 9... -0.43"]

1. e4 {+0.00/72} c5 2. Nf3 {+0.00/72 0.1} Nc6 3. d4 {+0.00/72 0.1} cxd4 4.
Nxd4 {+0.00/72 0.1} Nf6 5. Nc3 {+0.00/72 0.1} d6 6. Bg5 {+0.00/72 0.1} e6
7. Qd2 {+0.00/72 0.1} Be7 8. O-O-O {+0.00/72 0.1} O-O 9. f4 {+0.00/72 0.1}
Nxd4 {-0.43/7 4}

[Annotator "7. -0.20 1... +0.00"]

1. d4 Nf6 {+0.00/72 0.1} 2. c4 e6 {+0.00/72 0.1} 3. Nf3 b6 {+0.00/72 0.1}
4. g3 Ba6 {+0.00/72 0.1} 5. b3 Bb4+ {+0.00/72 0.1} 6. Bd2 Be7
{+0.00/72 0.1} 7. Nc3 {-0.20/6 4}
https://rwbc-chess.de

trollwatch:
Chessqueen + chessica + AlexChess + Eduard + Sylwy
Daniel Shawul
Posts: 4185
Joined: Tue Mar 14, 2006 11:34 am
Location: Ethiopia

Re: Scorpio MCTS+NN: What are the strongest nets?

Post by Daniel Shawul »

@Guenther, I have no idea why it didn't save the book info in the first game. I used xboard to play these games.
I could have used short opening lines (instead of native books) but tscp is kind of hard to setup in cutechess-cli without a book.
I am sure it is possible I just didn't try enough.

@Brendan Generally you should be using scorpio 2.8.8 not 2.8.7 as that have fixed multi-core CPU issues and ability to use multiple GPUs as well (though not efficient). The egbbdll64.dll i have up on my webiste is not optimized (only have sse2) but it works for almost everybody. I will upload different versions of it with AVX/AVX2 + FMA that will be significantly faster for people with recent hardware. I will post here once I do that.
User avatar
Tibono
Posts: 79
Joined: Sat Aug 01, 2015 6:16 pm
Location: France

Re: Scorpio MCTS+NN: What are the strongest nets?

Post by Tibono »

Hi Daniel,
thanks for sharing Scorpio MCTS.
I run it fine using epd NN (downloaded from Github: Scorpio/releases), but could not find ccrl NN?
Both links to NN in your home page seem to be currently broken.
Regards, Tibono